zsh-ollama-suggest

Command suggestions for Zsh using Ollama. Uses local LLMs to suggest commands based on your input, history, and current directory.

Features

• Suggests commands as you type
• Uses command history and directory contents for context
• Processes suggestions asynchronously
• Two modes: realtime (as you type) or manual (on demand)
• Configurable settings for model, suggestions, and context

Prerequisites

• Zsh shell
• Ollama installed and running
• git for installation
• curl for API requests
• jq for JSON parsing
• ls, awk, tail, head, grep for directory and text processing
• zsh-async (included as submodule)

Installation

Using Oh My Zsh

1. Clone this repository into your Oh My Zsh custom plugins directory:

git clone --recursive https://github.com/realies/zsh-ollama-suggest.git ${ZSH_CUSTOM:-~/.oh-my-zsh/custom}/plugins/zsh-ollama-suggest

2. Add the plugin to your .zshrc:

plugins=(... zsh-ollama-suggest)

Manual Installation

1. Clone the repository with submodules:

git clone --recursive https://github.com/realies/zsh-ollama-suggest.git

2. Source the plugin in your .zshrc:

source /path/to/zsh-ollama-suggest/zsh-ollama-suggest.plugin.zsh

Usage

1. Start typing a command (suggestions appear after 2 characters)
2. Suggestions will appear below your input:
   • Automatically in realtime mode
   • On demand with Ctrl+X c in manual mode
3. Navigation:
   • Up/Down arrows to cycle through suggestions
   • Ctrl+C to clear suggestions and reset
   • Ctrl+X t to toggle between realtime/manual modes
   • Ctrl+X c to manually trigger suggestions

Configuration

The following variables can be set in your .zshrc before sourcing the plugin:

# Ollama model to use (default: llama3.2:3b)
typeset -g ZSH_OLLAMA_SUGGEST_MODEL='llama3.2:3b'

# Ollama server URL (default: http://localhost:11434)
typeset -g ZSH_OLLAMA_SUGGEST_URL='http://localhost:11434'

# Maximum number of suggestions to show (default: 5)
typeset -g ZSH_OLLAMA_SUGGEST_MAX_SUGGESTIONS=5

# Number of history entries to consider (default: 1000)
typeset -g ZSH_OLLAMA_SUGGEST_HISTORY_SIZE=1000

# Model temperature for suggestion diversity (default: 0.1)
typeset -g ZSH_OLLAMA_SUGGEST_TEMPERATURE=0.1

# Number of directory entries to show in context (default: 25)
typeset -g ZSH_OLLAMA_SUGGEST_DIR_LIST_SIZE=25

# Suggestion mode: 'realtime' or 'manual' (default: realtime)
typeset -g ZSH_OLLAMA_SUGGEST_MODE='realtime'

# Debug logging (default: 0)
typeset -g ZSH_OLLAMA_SUGGEST_DEBUG=0

Debugging

If you encounter issues:

1. Enable debug mode by setting ZSH_OLLAMA_SUGGEST_DEBUG=1 in your .zshrc
2. Debug logs will be written to /tmp/zsh-ollama-suggest.log with millisecond timestamps
3. Ensure Ollama is running and accessible at your configured URL
4. Verify your model is downloaded (ollama list)
5. Check the log file for detailed operation tracing

License

MIT License - see LICENSE for details.

Acknowledgments

• Ollama for the local LLM runtime
• zsh-async for async processing
• zsh-autosuggestions for inspiration